Agento: The missing dashboard for Claude Code Agento, a new open-source dashboard from Shaharia Lab, turns Claude Code's session files into cost analytics, productivity insights, and a searchable history, with a browser UI for building and scheduling agents. The single Go binary requires no API key, account, or telemetry, and installs via `brew install shaharia-lab/tap/agento` followed by `agento web`, opening at `http://localhost:8990`. It attributes costs to the actual model used, tracks metrics like cache hit rate and tool error rate, and supports filtering and paging in SQL for scalability. Claude Code forgets everything the moment it exits. Agento reads the session files it already writes to your disk and turns them into cost analytics, productivity insights and a searchable history of every run. It also gives you a browser UI for building agents, scheduling them, and connecting them to the tools you use. One Go binary. No API key, no account, no telemetry. Your history, your agents and your analytics stay on your machine. brew install shaharia-lab/tap/agento agento web That is the whole setup. Agento opens at http://localhost:8990 , finds your Claude Code history, and starts building your dashboards. If Agento saves you money or time, a star is the single most useful thing you can do for the project. It takes a second and it is how other Claude Code users find it. Requirements: the Claude Code CLI https://claude.ai/code , installed and authenticated. If claude runs in your terminal, Agento works. No Anthropic API key is needed, because Agento uses the authentication Claude Code already has. | brew install shaharia-lab/tap/agento agento web | grab the archive for your platform from Releases tar -xzf agento Linux x86 64.tar.gz sudo mv agento /usr/local/bin/ agento web | Binaries for Linux x86 64, arm64 , macOS Intel, Apple Silicon and Windows are on the Releases page https://github.com/shaharia-lab/agento/releases . Useful flags: agento web --port 3000 to change the port, --no-browser to skip opening a tab. To keep it running in the background across reboots: agento service install then: status | stop | start | restart | logs | uninstall Input, output, cache reads and cache writes bill at very different rates, so Agento keeps them apart instead of multiplying one total by one price. The result is the chart most people find surprising: the model with the most tokens is often not the model taking your money. Cost is attributed to the model that spent it, including work done inside sub-agents, so delegating to a cheaper model shows up as an actual saving. The Insights page goes past raw counts. It tracks how many turns a session needed, how far Claude got before it had to ask you something, how long you kept it waiting, cache hit rate and tool error rate, all compared against the previous period so you can see the direction. It then attributes every tool call to the skill, plugin, MCP server or sub-agent responsible, which is how you find the skill quietly burning a third of your calls. Durations mean active time, not wall clock. Claude Code sessions are resumable, so one picked up a week later would otherwise report a week of work. Idle gaps beyond a threshold you control are excluded everywhere a duration is shown. Sessions per day, model mix, busiest days, cost per project, and an activity heatmap that counts a session in every hour it was running rather than only the hour it finished. Filtered and paged in SQL, so it stays fast whether you have 50 sessions or 5,000. Search across titles and content, filter by project, model, date, cost or duration, and see linked pull requests, git branch and permission mode on every row. The journey view reconstructs the full timeline of a run: every prompt, response, tool call and result in order, with each sub-agent's steps nested under the delegation that spawned it. When a long autonomous run goes wrong, this is where you find out where. Each session also carries its own metrics: turns, steps per turn, longest autonomous chain, active duration, time Claude spent working, your own average reply time, and tool error rate. Give an agent a name, a system prompt, a model, a thinking mode and an explicit list of tools it may use. Save it once and reuse it from the browser, the scheduler or the CLI. Template variables like {{current date}} are filled in at runtime. Permission modes matter here: an agent can be set to plan first and act only after you approve, which is what you want the moment it can write files or run commands. Run any agent on a cron expression, a fixed interval, or once at a specific time. Every execution is recorded with its status, duration and full output, so you can see exactly what ran while you were away. Each integration runs as an in-process MCP server, so there is no extra daemon to operate. Configure it once and any agent can use it. Google Calendar, Gmail, Drive , GitHub, Slack, Jira, Confluence, Telegram and WhatsApp are built in. Any other MCP server can be added through ~/.agento/mcps.yaml over stdio, streamable HTTP or SSE. Rates ship for Anthropic, Moonshot, Z.ai and Alibaba models, and they are effective-dated: adding a rate leaves past usage priced at what it was charged, while correcting one rewrites it. A model with no published rate is reported as unknown instead of being quietly priced as something else. Agento reads ~/.claude and caches results in a local SQLite database at ~/.agento/agento.db . Nothing is uploaded, there is no account, and there is no server component. Projects you would rather leave out of the numbers can be hidden from every report, and the idle threshold behind the duration metrics is yours to set. 💬 Chats and a tabbed multi-chat workspace Hold multi-turn conversations with any agent you have built. Responses stream live over Server-Sent Events, sessions persist locally, and you can favourite or rename them. Drag and drop files or paste images straight into the input. The multi-chat workspace runs several conversations in parallel, each tab with its own agent and session state, and it survives a page reload. demo.webm Multi-Tab.Chat.-.Demo.webm 💻 CLI: run agents from the terminal agento ask "What changed in the repo today?" agento ask --agent code-reviewer "Review the staged diff" agento ask --agent code-reviewer "Follow up"